NSF 06-501: Frequently Asked Questions for High Performance Computing
System Acquisition: Towards a Petascale Computing Environment
for Science and Engineering (NSF
This page provides clarification and pointers to additional material
that might be of interest to people and institutions considering
submitting a proposal in response to the solicitation named above.
The full text of the solicitation may be found at
- What are some additional
sources of information that might be helpful to prospective resource
Prospective resource providers might find the following to be
useful sources of background information:
Information about usage of existing NSF-funded major HPC centers.
http://www.sdsc.edu/user_services/allocations/past/ - LRAC, MRAC
and NRAC allocations
Resource allocations at NCAR’s Climate Simulation Lab.
General information about Teragrid
http://www.teragrid.org/ - TeraGrid home-page
Information about Teragrid and how a center’s personnel
might participate in TeraGrid activities.
Information about NSF awards
http://www.nsf.gov/awardsearch/ - search for project abstract by
investigator name, project title or award number.
- How will the benchmark data be used?
All of the proposal contents, including actual or estimated
benchmark data included with the proposal, will be provided to
reviewers. Reviewers will also have access to a copy of the solicitation
and to information about the benchmarks that proposers were asked
to run. Reviewers will be asked to evaluate proposals based on
consideration of both the qualitative and quantitative information
supplied in the proposals. NSF will consider both the proposals
themselves and the reviewers’ evaluations of the proposals in selecting
proposal(s) for award. NSF’s decision-making will also
take account of both the quantitative and qualitative information
in the proposal. NSF views the benchmark data as information
that is important but not the sole determinant in funding decisions.
As indicated in the solicitation, performance indicated by benchmark
results may be used as the basis of performance measures included
in award documents as acceptance criteria or other conditions of
- Does the limit of two proposals mean to each deadline date
or to all of the deadline dates as a whole?
An organization may submit up to two proposals to each annual
competition (or deadline date). Thus, an organization may submit
up to two proposals on or before February 10, 2006 for consideration
in the FY 2006 competition, and an additional two proposals on
or before November 30, 2006 for consideration in the FY 2007 competition,
and so on.
- If I am interested in proposing the acquisition of a highly
experimental computing system, will it fit within this solicitation?
The emphasis of this solicitation is on providing resource(s)
that maximize scientific and engineering research productivity
in a broad range of areas. Proposals for systems for which the
risk of failing to meet this goal seems unacceptably high, to the
reviewers or to NSF, are not likely to be competitive.
- Proposing organizations are required to ensure open access
for researchers; what does this mean?
Awardees will be required to provide system access to personnel
associated with projects that receive HPC allocations through the
national Resource Allocation committees, the Large Resource Allocation
Committee and the Medium Resource Allocation Committee. This is
not intended to curtail the right of an awardee to restrict access
for personnel who have misused their access privileges. Nor should
it be interpreted as requiring awardees to provide users with physical
access to the HPC systems.
- The solicitation says that, “Proposals that request
support for HPC research will be deemed ineligible and returned
without review.” What is meant by “HPC research”?
Awards made under this solicitation are intended to fund the
acquisition or upgrade and subsequent operation of an HPC system.
Proposers should not request funds to directly support research
activities, such as 1). research where high performance computing
is the subject of investigation or 2). scientific research that
requires access to high performance computing systems. Funding
for such research should be sought via proposals submitted to
and education programs.
- What is involved in making HPC resources available
via the TeraGrid?
TeraGrid is a collaborative infrastructure consisting of resource
providers (RPs) and a Grid Infrastructure Group (GIG). The GIG
is responsible for overall TeraGrid architecture, planning, and
management, as well as for providing and/or coordinating common
and central services. The TeraGrid system itself is an integrated
and coordinated set of resources that provide advanced capabilities
to users, science and engineering researchers and educators.
The system design and implementation are driven by the requirements
of scientists and engineers, and the system’s functionality
is delivered through a variety of software, middleware, policy,
and support functions.
Integrating and operating a computational resource as part of
the TeraGrid involves a set of required and optional
activities and services. Required components include:
for TeraGrid data movement services;
- participation in the
coordinated implementation of security practices and policies;
in problem resolution for issues related to local resources;
for the Coordinated TeraGrid Software and Services specification;
in verification and validation processes; and
in the resource allocation and accounting processes.
The experience of current TeraGrid sites
suggests that the initial task of integrating a fully operational
computational resource into the TeraGrid system involves 6-10
weeks of work by 4-6 staff members with appropriate expertise. For resources that involve
novel architectures or operating systems the initial integration
and ongoing maintenance may be more involved. The GIG includes
staff members who can assist in the integration effort, including
on-site assistance if needed.
The nature of providing a computational resource for the national
user community requires a significant effort in areas such as networking,
security, software maintenance and system administration, and accounting
and allocations. Ongoing participation in TeraGrid need not add
substantially to the base level of staffing necessary to support
a national user community. It
is estimated that participating in TeraGrid will require
approximately 1 FTE in additional effort, primarily
in supporting TeraGrid middleware functionality and
in security, accounting, account management, and allocations
coordination. In addition, participating sites will
need to provide the GIG with responsive points of contact
for system administration, networking, security, operations,
and user support. However, more active participation
in TeraGrid is also welcome.
For coordination and planning purposes, TeraGrid uses working
groups (ongoing) and requirements analysis teams (specific short
term planning efforts), open to all TeraGrid partners who wish
to participate. TeraGrid support, software planning, and security
policy are examples of the activities covered by such groups. Additionally,
TeraGrid partners hold a weekly Access Grid meeting (standard telephone
call-in is also supported) to discuss upcoming plans and ongoing
activities. With the exception
of the security coordination group, participation in
these discussions is optional, however participation
benefits both the individual staff members and the
RP institution. Staff members benefit through involvement with peers who have similar
or complementary experience and skills. Institutions benefit because
these activities provide an opportunity to help set policy and
technical direction for TeraGrid. The TeraGrid activity itself
benefits through the engagement of a broad range of perspectives
The TeraGrid partnership has prepared a primer on the more general
aspects of TeraGrid from the perspective of a Resource Provider.
This may be found at http://www.teragrid.org/basics.
- How will maintenance, operations and user support be funded
and what budgets should I include?
As described in detail in Section IV (Award Information) of the
solicitation, the anticipated funding amount of up to $30,000,000
is intended to cover the costs of acquiring, installing and bringing
to an operational status an HPC system or systems. The primary
budget entered on the normal FastLane budget pages should be for
A separate budget for user support and operating costs, as described
in the section, “Effective User Support and Projected Operating
Costs,” in Part V.A of the solicitation, should be included
in the Supplementary Documents section of the proposal.
Reviewers will be asked to consider the user support and operating
costs in their analyses of the proposals and these costs will be
a factor in funding decisions. At the time that the terms of an
award are being negotiated with a prospective awardee, NSF will
also negotiate a budget for user support and operating costs and
ask the prospective awardee to submit a supplemental budget to
cover these costs.
- What sort of information might help me
to estimate the mix of jobs that is likely to represent the usage
of the system?
In addition to a proposing organization’s experience as
a provider of HPC services, material on current usage of resources
at NSF-funded systems is available at some of the web-addresses
listed in FAQ 1. A number of research communities have articulated
their future HPC science and engineering goals in workshop reports
and similar documents. A number of these may be accessed from
the OCI web page, http://www.nsf.gov/div/index.jsp?div=OCI.
- Will additional FAQs be added and how do I find out when
this has happened?
If questions are received from prospective proposing organizations
that seem likely to be of general interest, they and our answers
will be added to this set of FAQs. We do not know if and when
this will happen so please check this page frequently for updates.
- Is it permissible to include, in a single proposal,
options for a particular architecture deployed at two or more
different scales at correspondingly different costs? E.g. Option
1: system X for $YM; Option 2: 2 times System X for $ZM.
No. This will be interpreted as proposing a choice
between multiple systems within a single proposal. The solicitation
states that, "Each proposal should be for the acquisition
and deployment of a single HPC system for which the project
costs total between $15,000,000 and $30,000,000."